Picture orientation information in video compression

This paper proposes changing of picture orientation on a picture-by-picture basis, without requiring the encoding of an Instantaneous Decoder Refresh (IDR) picture. It can be implemented by small modifications to H.264/MPEG-4 AVC and the design of the upcoming High Efficiency Video Coding (HEVC) standard, or most other video coding technologies employing inter picture prediction. To enable picture orientation changes, a high layer syntax element is needed to indicate the rotation to be applied to the decoded pictures. Modifications to the decoding process are also required to avoid the use of IDR pictures. For some applications with handheld cameras that may be held at different orientations, the proposed modifications allow shifting the rotation function to the decoding end, reducing overall system complexity. Additionally, selection by the encoder of the coded picture orientation may result in better coding efficiency. In this respect, an algorithm for selecting the coded picture orientation limited to intra coding is also described, and experimental results show coding efficiency improvements up to 3.4% BD-rate gain.